Module aggregate 

Cross-run aggregation, the “accurate averages” layer.

A benchmark cell is one CellKey: a scenario name with the whole topology it ran at. The harness runs each cell N times per stack. This module reduces those N RunOutputs into a mean ± sample-stddev per metric, and averages the per-interval time series across runs at each time offset. Both the Markdown summary and the Plotly graphs read these aggregates, so one place computes “the average” and no renderer re-derives it.

Structs

CellAgg

One benchmark cell. It holds a (scenario, topology) with the krabka and kafka aggregates side by side.

CellKey

What makes two runs comparable: one scenario at one topology.

ScalarMetric

A headline scalar metric. It holds its key, a way to pull it from a run, and which direction counts as “better”, so a renderer can colour and ratio it correctly.

StackAgg

Per-metric aggregates for one stack within a cell.

Stat

Mean and spread of a single scalar metric over the runs of one (cell, stack).

TsPoint

One averaged point of a time series. It holds the across-run mean of a metric at a fixed offset into the run, and how many runs gave that offset.

TsSeries

An across-run-averaged time series for one (cell, stack, metric).

Functions

aggregate_cells

Groups runs by CellKey and reduces each stack’s runs to a per-metric Stat. This returns the cells in key order.

averaged_timeseries

For each (cell, stack, metric), averages the per-interval samples across all runs at each shared time offset.

scalar_metrics

The headline scalar metrics rendered as krabka-vs-kafka bars.

stat

Mean, sample stddev, and count over a sample of run-level values.